레시피에 따른 요리 영양소 예측 신경망 모델

Abstract

Although food science and technology have advanced, predicting nutrient changes during the cooking process remains a challenging issue. This study proposes a method for automatically predicting nutrient changes caused by cooking processes using a nonlinear regression neural network. Chemical changes during cooking significantly impact nutrient content, such as carbohydrates, proteins, and fats, which are critical for personalized diet planning and nutrition management for individuals with chronic diseases. To address this, natural language processing (NLP) techniques were applied to align ingredient names across differently formatted datasets, leveraging BERT and Food2Vec models to refine and merge the data. Using datasets from Food.com and USDA, relationships between ingredients and cooking methods were quantitatively represented as vectors to develop a model capable of predicting the nutrient composition of dishes. The first model employed a simple neural network structure but faced limitations due to data scale issues. The second model addressed these limitations by applying logarithmic transformation to normalize data distribution. The final model combined log-transformation with residual learning to effectively capture the nonlinear characteristics of the data while minimizing information loss. Experimental results showed high prediction accuracy for certain nutrients such as saturated fats and proteins, although limitations were observed in underestimating or overestimating extreme values for carbohydrates and sugars. This study demonstrates the potential of a neural network-based model to automate the prediction of nutrient changes during cooking processes, offering practical applications for personalized diet planning and advancements in the food technology industry.